LAUSR

Hyperspectral images (HSIs) have fine spectral information, and rich spatial information, of which the feature quality is one of the key factors that affect the classification performance. Therefore, how to extract essential features, and eliminate redundant features from hyperspectral data are the main research focus of this article. Here, we propose a spectral-spatial feature extraction method based on ensemble empirical mode decomposition for HSI classification, which contains several steps as follows: First, the dimension reduction for HSI is performed by using the principal component analysis method. Second, in order to decrease the sensitivity to noise, and extract rough outline features, the adaptive total variation filtering (ATVF) is conducted on the selected principal components. Furthermore, by using the ensemble empirical mode decomposition (EEMD) to resolve each spectral band into sequence components, the features of HSIs can be better coalesced into the transform domain. Finally, the first $K$ principal components of the input image, and the outputs of the ATVF, and EEMD are integrated into a stacking system to obtain the final feature image, which is then classified by a pixel-wise classifier. The experimental results of three authentic hyperspectral datasets show that the proposed algorithm obtains a superior classification performance compared with other methods.